A significant weakness of the analyses performed at low resolution is the inability to study recombination on a level of the individual functional units, meiotic hotspots. In contrast, high resolution analysis has the power to assess the functional activity of individual hotspots. Two approaches are frequently used to define profiles of recombination at high resolution. The first approach is the reconstruction of high resolution recombination rate profiles from genetic variation data. The second approach, mentioned above, is sperm genotyping.
The first approach uses the genetic variation found in populations to estimate recombination rates. What is now seen as single nucleotide polymorphisms (SNP, the most frequent kind of sequence polymorphism) was at some time in the past a rare mutation in an ancestral chromosomal context. These ancient chromosomes have been mixed in thousands of generations to produce present-day chromosomes, but the historic association of SNPs is still detectable. These patterns of non-random association of genetic markers, called linkage disequilibrium (LD), are shaped by past recombination events [33
]. Computer modeling allows reconstructing the population history backwards in time and estimating the likelihood of a given recombination frequency between pairs of polymorphic markers. An ever increasing number of methods have been developed for the estimation of recombination rates from population variability data, the most accurate of which are based on coalescent reconstruction [for review see 34
]. One potential drawback with many of these methods, in particular those that are more accurate in estimating rates, is the extreme computational demands. Full coalescent reconstruction is absolutely prohibitive from a computational point of view. Several approximations to the calculation of full-likelihood have been introduced (two programs, LDHat [36
] and Phase [37
], are most popular, and see [34
] for discussion). Although these approximate methods are still computationally-intensive, rapidly increasing performance of modern computers makes it possible to perform such calculations even on a genome-wide scale and such methods have been applied successfully to calculate high resolution recombination rate maps of the human genome [8
]. The advantages of computational methods are speed, throughput and cost-effectiveness. It is relatively easy to collect DNA samples from 30–50 individuals and determine their genotypes using modern genotyping techniques. These genotype data can then be used to calculate recombination rate profiles.
The variation between population-specific recombination rate profiles has been documented in several studies [8
]. The variation in calculated profiles of recombination was first comprehensively studied by Clark and coauthors [40
]. They found evidence of population heterogeneity at ~100-kb resolution in many of the 538 SNP clusters studied [32
]. Later studies were performed at higher resolution allowing analysis of individual hotspots. For example, by performing a computational analysis of 74 genes resequenced in 47 individuals as a part of the SeattleSNP program (http://pga.gs.washington.edu/
), Crawford et al. found evidence of hotspots in 35 genes and in 16 of these 35 genes (45%) a hotspot was found only in one population [42
]. A more sensitive recent analysis of a slightly larger set of SeattleSNP genes again found that 35% (43 out of 121) of all hotspots were detected only in one population out of two [43
]. What has been missing from studies performed so far is an accurate unbiased estimate of the statistical significance of the observed differences on a genome-wide scale.
An interesting application of the LD-based computational methods is the comparison of profiles of recombination in closely related species [46
]. In these studies the authors have reconstructed recombination rate profiles in the orthologous regions of the genome in human and chimpanzee based on population surveys of genetic variation. Despite a 98.7% identity in DNA sequence, there is no correlation in the positions of hotspots in up to 14 Mb of sequence [46
]. Thus, hotspots of meiotic recombination have completely changed their positions in the ~7 MYR since the split between the human and chimpanzee lineages. At the same time, in an apparent contradiction to previous findings a recent study has shown that hotspots are found in the same location in paralogous genomic loci with an age of duplication preceding the human-chimpanzee split [50
]. It is unclear, however, how general is such a conservation in hotspot position.
An intrinsic problem associated with the computational methods discussed is the fact that they are based on calculating population-averaged recombination rates from a sample of individuals. Thus, measuring the individual-specific recombination activity is impossible in principle. Another caveat is the fact that the inaccuracy in defining recombination rates from sequence variation data using coalescence-based approaches is rather high [8
]. Thus, it is likely that many or even the majority of the observed differences between population-specific recombination rates do not reflect true biological variation. In addition, the definition of confidence intervals of recombination rate estimates with respect to true biological rates is not trivial [36
]. Another potential problem is that the majority of coalescent-based methods are based on a very simple model of DNA recombination that assumes that hotspots are completely conserved in a population. We now believe that this is an oversimplification and there is a great deal of variation in hotspot strengths among individuals. Recently introduced methods try to address this issue of the possible variability in hotspots and incorporate the heterogeneity explicitly in the likelihood calculations [52
], however, this is still an area of active development. The development of more realistic recombination rate models that incorporate hotspots explicitly, as such [54
] will also improve the quality of predictions and the power of analysis.